Predicting railroad crossing crashes
Every year in the United States, more than 200 people lose their lives at railroad crossings. Although the number of crashes has been steadily declining in recent decades, the result of a vehicle–train collision is often catastrophic. To ensure the number of rail crossing crashes continues to decline, it’s important for highway departments to invest in safety improvements at the locations where those improvements will have the greatest impact. However, the models currently used to predict where rail crashes will occur are often imprecise.
“Today, most highway departments rely on the USDOT model to predict rail crashes, which was developed using data from the 1974 database, and most of the coefficients remain unchanged since 1980,” says Rahim Benekohal, a professor in the Department of Civil and Environmental Engineering at the University of Illinois at Urbana-Champaign and researcher with the Roadway Safety Institute.
At an October 2 Roadway Safety Institute seminar, Benekohal described a project that aimed to improve on the USDOT model by developing a more accurate crash-prediction model for rail crossings. Benekohal and his research team created the new model by applying a macroscopic approach, which traditionally identifies general trends in national or regional data, to data from Illinois. The researchers also incorporated variables identified in a detailed micro analysis—which examined individual crashes at high-crash locations to identify contributing factors—into the macro approach.
To complete the micro analysis, the researchers looked at several high-crash crossings, first completing a detailed crash diagram of each crossing to visualize key information such as where the crashes happened and what types of crashes occurred. Next, they created a dynamic tree structure for each location to identify crash trends. In one instance, this process revealed that the intersection angle was contributing to a number of crashes. At a second location, they found all crashes involved elderly drivers—likely due to the high number of assisted-living communities in the area.
“By performing this type of analysis at the micro level, we can identify contributing factors that will enable highway departments to select the proper countermeasures at individual high-crash crossing locations,” says Benekohal. In addition, the researchers can take those factors and determine whether they should be included in a macro-level model. “For example, we found that angle was a significant contributing factor to rail crossing crashes. So in our new model, we use crossing angle as one of the variables to predict the number of crashes at a crossing.”
The early results of the project are promising, Benekohal says. The new crash model is already more accurate at predicting the number of crashes at rail crossings and ranking high-crash locations than the USDOT model, and researchers hope additional improvements will provide an even greater level of precision.
Currently, Benekohal’s research team is continuing to refine the new model with funding from the Roadway Safety Institute. The researchers will also study train delays to estimate train arrival times at rail crossings, develop tools to support emergency response planning and investment, and develop models to optimize the coordination of emergency response to railroad crashes across jurisdictions.